UAV4GEO Thermal tool question

Alessio

Good morning,
I am a second-year Master's student in Environmental and Land Engineering at the Polytechnic University of Turin. For my thesis, I am processing thermal images captured with a DJI drone and using UAV4GEO’s Thermal Tool to convert them from R-JPG to TIFF format.
I am writing on behalf of myself and my supervisors to seek clarification regarding some issues encountered during the thermal image processing—specifically concerning the environmental parameters entered into the tool. When setting these parameters, we noticed that the data acquisition distance cannot exceed 25 meters (a limitation also observed when using the DJI Thermal SDK, which shares the same conversion model); however, the drone used for our project flew at much higher altitudes, often exceeding 200 meters. Since the Thermal SDK also specifies a 25-meter distance limit—instructing users to input this value even if the actual acquisition distance exceeds 25 meters—we were wondering if there is a way to enter a value higher than this imposed limit (notably, the error screen appearing in the tool when entering parameters like emissivity = 0.1 or reflected temperature > 100°C indicates a distance range of 1 to 300 meters). Alternatively, if this is not possible, we would like to know what impact setting the value to 25 meters has on the quality of the resulting conversion.
Thank you very much for your assistance.
Alessio Calì

51 Replies

dancejod

Hi Alessio! What sensor and SDK version are you using? I might be able to share my experience as I have been through this myself recently

EDIT: apologies I did not check the SDK version in ThermalTools. I will continue with my reply in a bit

dancejod

For my thesis, we flew with H20T at 40 meters AGL. We were curious about the exact same thing. We were capturing a peatland patch and fortunately there were some water ponds, so I cropped one of our images and extracted a homogeneous water patch of 50x50 pixels, ensured the emissivity is set to 1 and the parameters humidity and reflection set to default, and only changed the distance during the conversion to tiff. Here is how the temperature of the patch changed. The response is nonlinear.

It seems that the distance parameter can indeed be a problem for some applications requiring high measurement accuracy. I tested this out on version 1.18 of the SDK (I used an old one 1.15 in my thesis) and it seems that the 25 m distance cap has not been resolved yet. I remember looking for papers on this, and I think at least this one acknowledges the cap.

Note that the response is from the 40 m AGL flight and it might be different for your 200 m flight, I'd be curious to see it!

For such high flights, based on the discussions with other folks in academia who were doing a similar research, it seems that it would be better to stay away from the proprietary DJI thermal cameras for high accuracy demanding applications... If you care more about the relative temperature patters I suppose it is a lesser deal.

Let me know if I can be of any more help!
Daniela

👍 1
Alessio OP

Thanks a lot for te answer. I tried converting the thermal images we acquired with UAV4GEO setting the distance at 25 m but when i tried to open them on QGIS to visualize the result i noticed that the converted images had a wrong coordinate metadata (the images had no projection set and when i specified the reference system they were all overlapped. I tried opening the metadata of the converted images and the coordinates are correct but in QGIS they are not applied) and also the number of pixel in the converted image has been halved (the original image was 1280x1024 while the converted one was 640x512). Did you or someone else have the same problem?

pierotofy

I think you're expecting the tool to georeference/reproject the thermal images as GeoTIFF, but that's not what the tool does, ThermalTools extracts the celsius values (which are 640x512) to a standard 32bit TIFF and copies the EXIF meta from the original image. If you want to see the results on a map, first process the images with WebODM.

Alessio OP

Thanks for the clarification, insetad for the cap of 25 m tò the acquisition distance Is there a way to raise the distance value? And of not how does the distance impact on the result of the conversion with a distance of acquisition that is much higher than the limit (we flew at around 200-250 metres)?

dancejod

The distance 25 m is unfortunately the capped maximum. I am not familiar with the SDK binaries and maybe it would be possible to tinker with it, but I am not sure if it would be a right direction. I think all we can do with DJI thermal sensors is just to simulate what the distance parameter does to the measurements. Once upon a time I was reading through some forums to see if someone had managed to reverse-engineer the equations for H20T's radiometric calibration but I didn't find anything...

Honestly I am not sure why it is capped to 25 m. Could it be that they had some specific industrial applications in mind?

I am however going to run some experiments; I have some more datasets of the same study area taken at 50 m AGL and 100 m AGL. I will take a look at what the distance parameter does to the water patches taken from these altitudes and share it here. It should change things a bit as the reflected radiance has to travel a longer path through the atmosphere to reach the sensor

GenAnon

FWIW I wonder if the thermal cap of 25m is a matter of calibration and manufacturing production. These thermal sensors are affected by ambient environment and sensor sizes. Not unlike regular rgb camera sensor size dictates the megapixel value and how much light is possible to capture. Both sensors can take photos from any distance but pixel value and resolution is limited by sensor sizes.
With thermal the resolution is already much less than an rgb camera in the same machine and determining accurate radiance is impacted. Then processing the captured results to models and maps is a matter of sfm/ODM doing the best it can on lower resolution pixels.
Also I was once told for these cameras to effectively detect thermal difference the subjects need to have a certain temperature difference for the sensor to detect thermal difference. If a subject (person) is 98deg f and the ambient environment (ground) is also 98deg f then the sensor detects all of the radiance as the same value. That could also be an impacted factor with 25m cap and results.

👍 1
Winux Nomacs Best answer

@GenAnon you may want to try the DJI Thermal Analyzer plugin just released as it allows it use the laser rangefinder distance in the converted photos.

🤘 1 🔥 1
Alessio OP

Thank you, can this plugin convert the thermal images into TIFF format or it can only leave the format to JPG? Because i installed it but i can't understand how the conversion and exportation works

Winux Nomacs

The plugin converts the images to tif format in order to be recognized as radiometric images when processed by webodm. The export feature of the plugin allows you to choose your desired export format whether source, tif, jpg or basic png for visible light representation only.

Alessio OP

I tried a coversion but i couldn't find how to convert to tif and export the image in that format, can you tell me where can i set the export format?

24Kerrick

Follow up, what sensor are you using the current M4T SDK does allow 1 to 300m change, the Thermal Tools Piero made I am sure has to do with a limit on the hardware.

What is your end goal outside the conversion? The DJI Thermal CLI tool Winux made does a great job to mosaic images, then let you reference the originals, no photogrammetry needed.

Alessio OP

I Just realised i was looking at a different tool than the one that was mentioned before, my fault.
The end goal should be the creation of a mosaic and other analysis for my master thesis, i still don't know very well the final goal because i'm working with a research team and they are still understanding what is better to analyze

24Kerrick

Are you looking for a true orthophoto (photogrammetry) with ground control? or a stitched mosiac (non-photogrammetry). I would say give his tools for outside webODM a try, I was able to get really clean results, has some cool features, he would be able to help you more.

If that doesn't work come back we will get you taken care of.

A big part of this will be trying to understand your end goals to help with the capture + processing, if we didn't capture well, processing would be hard. Knowing your equipment will help as well.

Winux Nomacs

The uav4geo thermal tools will create a folder of converted images that you then have to import as your source images into WebODM. Unless you use the DJI Thermal Analyzer plugin for WebODM (built upon the dji-thermal-cli) which upon creating a thermal task, handles the conversion for you. https://github.com/WinuxNomacs/webodm-dji-thermal

Winux Nomacs

Install the dji thermal analyzer > Create Thermal Task > Select your source m4t files > Start Processing. These are the initial steps you'll need to follow to generate the Orthomosaic in WebODM, full photogrammetry via ODX

Shawn (River Valley Aerials LLC)

Are there plans to do any of the other thermal drones? I work at a school we have a skydio X10.

pierotofy

Always looking to add support for more drones; if you can share a small dataset (~20 images) it would help expand support

Shawn (River Valley Aerials LLC)

what images would you like? This photo shows the file structure from the SD card.

pierotofy

Upload them somewhere like Google Drive or Dropbox

GenAnon

Just upgraded my “day job”laptop version of WebODM to latest .exe release. Found the DJI Thermal Analyzer plugin and installed it—no issues for those steps.
Next up: test of thermal images captured with M30T. We flew an earlier mission which captured all cameras. I’m curious now to see what this will do.
I’ll give it a test with older M2ED thermals but that probably won’t work out because they didn’t generate radiometric output then.
Will also test with M4T when we get a good collection.

Thanks for the plugin!

Winux Nomacs

That's awesome! I run a M4T myself. Hey if you do run into any issues, just holler. Also make sure you're using the latest, just released an update yesterday

GenAnon

Awesome to see results that can switch pallets and show temp under cursor!
65 images recognized as nadir 134 obliques (left out?). I’m digging this so far!
Flight wasn’t designed for great thermal overlap; focused on modeling in RGB. Great opportunity to add more capabilities with one machine!

Winux Nomacs

Remember the mosaics in the Thermal Analyzer are meant as preview. Idea being you can optimize your settings, view the preview and send them into WebODM processing for a real orthophoto. The mosaic generator by default leaves out shots where the gimbal angle is greater than -80 degrees.

👍 1
Winux Nomacs

You may just want to dm him that link mate. Unless you don't mind sharing with everyone.

GenAnon

Poke and hope! Learn and run tests with changes. Isn’t that what’s fantastic about WebODM?!—no SASS BS that charges me every time I want to tune and rerun.
That said, I just pushed the thermal task onto my project and ripped it! HERE WE GO 🚀
The result was only using the nadirs but sure is pretty so far!

24Kerrick

Looking good! Was this thermal only or did you run it with the daytime images?

Winux Nomacs

That's awesome, be sure to take a look at the Thermal Inspector hiding down there too

Winux Nomacs

And ty so much for sharing

Winux Nomacs

Ooooh I said a naughty word. Apparently we can't say peeved off

GenAnon

This was from thermal only. From the thermal analyzer I created the task back to the parent project. Then I set the processing node and let it run.
Tomorrow is a holiday so I’ll pick up this again Thursday. I’m generating the next task using all images.

GenAnon

Thank you.
Ran the thermal tools to generate new thermal output and 3D model. This time, I used analyzer to prepare and include the obliques in the task for the project. 65 nadir and 134 obliques = very pretty results.
One question for the 3D model is it possible to also switch palette and view temperature results?

I see the analyzer generated multiple copies of the source images (_T_temp_c.tiff; _f.tiff because I enabled that; and the radiometric jpg because I enabled that too).

Using gf show cameras on 3D will click to the _c.tiff.

Would be awesome to get radiometric renders of PC and textured model.

Is this possible and I didn’t find it yet?

Winux Nomacs

I might've been wrong @GenAnon I might see a way to do it though seam smoothing would make temp at the seams shift a bit but not bad, super local. I'll get back to you

Alessio OP

Just another question, i installed the Thermal Analyzer and tried to run a test conversion, but i have a doubt aboit the distances because i can select to use the embedded 5 m value, the distance measured by the sensor's laser or a fixed value. I tried running a conversion with all the three options and i noticed that with the embedded value i have temperature values close to the reference data that i have (coming from a meteorological station), while if i set the laser distance or if i set a distance value of any sort (i tried with 20, 100 and 200) i always find similar temperatures and in all the cases apart from the embedded value the results are way lower than the reference station. Can you explain me how the plugin manages the distance values? I ask also because i tried changing the distance values also on UAV4GEO (that allows to change it only in a range between 1 and 25 m) and what i found there was that temperature values were different with the 20 m setting, so i wanted to understand if the correction model used for the Thermal Analyzer was the same as UAV4GEO's (that if i'm not mistaken was the same of the DJI Thermal SDK)

👀 1
Winux Nomacs

Looking doable, I'm still validating the temps between the source, ortho and model to make sure the approach is viable.

👍 1
Winux Nomacs

So looking back, the 25m limit was based off of the H20T and supported models at that time. v1.8 of the sdk doesn't impose that limitation on the m4t. Let me dig into distance a bit more to give you a detailed response. It's actually how dji account's for moisture in the air but I don't remember how offtop

pierotofy

Yeah it depends on the drone; some models are limited to 25m, others have higher limits (up to 100m if I remember)

Winux Nomacs

m4t up to 300

👍 1
24Kerrick

Field test with something you know the emissivity of and play with the settings to see what works for you temp calc. I have thought about use black tape / foil as GCP's throughout a project. Would work as temp checks as well.

Winux Nomacs

Wouldn't a delta T on the ground just reflect the ambient radiation of the sky? Asking, I've not done this

Winux Nomacs

Yeah, crinkled foil on about a 12x12 is a Delta T Square used for calibrating the ambient radiation. Needed for thermography outside especially when the sun's up. Normally you'd place it vertical, next to your target to adjust your ambient/background temperature. If it's lying face up though it just reflects the sky

24Kerrick

Yup, the intent is how to do find a highly contrasting surface in thermal?

Winux Nomacs

So me and my pal Claude dug into it and compared my own test m4t projects at the same time. Essentially it would appear your background/ambient temperature may need some adjustment: The plugin passes the distance straight to DJI's Thermal SDK, the same correction UAV4GEO uses, so the model is the same. What you're seeing is that correction combined with the ambient temperature it's given.

The SDK removes the radiation the air between the camera and the ground adds. How much it removes depends on the distance, the air temperature and the humidity. Unless you type an ambient temperature, the plugin uses the one each photo records. That isn't a weather reading: the camera records its own estimate, the same value as the reflected temperature, and it drifts during the flight. If that value is warmer than the real air, a long air path makes the SDK take off too much, so everything reads low. Cooler surfaces drop the most.

On one of our M4T frames, with the recorded ambient temperature of 27.6 °C, the mean was 22.8 °C at 5 m, 21.8 °C at 21 m (the rangefinder's distance), 18.8 °C at 100 m and 17.8 °C at 200 m. At 100 m, setting the ambient temperature to 20 °C turned the frame's mean shift from −4.0 °C into +4.6 °C. Humidity barely mattered. This is why your 5 m results match the station while the longer distances all come out low: the 5 m path keeps the correction small.

So you can either:

  • Correct for the air: use the rangefinder's distance, or your flight height, and enter the ambient temperature and humidity your station measured at the time of the flight.

Leaving the air out: keep "As recorded" (5 m). This matches DJI's own apps. The frame page's Measure again shows what each value changes on a single frame before you convert the whole flight.

24Kerrick

Now with 20 throughout the project, next to things you want to measure the temp of. Would give you a check that you can use to verify a ground vs aerial check for calibrating radiometrically throughout the project for drift or changes in wind etc during capture.

Then you could try to "visualize it" based on that radiometrically.

Lastly, it gives you checkpoints to tie imagery together for alignment or possibly check absolute accuracy if you care

Winux Nomacs

Good luck and lastly remember to enable texturing-skip-global-seam-leveling for thermal tasks as well, or it skews the temperatures at the model seams

Winux Nomacs

Need more help?

This discussion happened in the WebODM community Discord. Join and ask your own question.

Join View Original Thread

This page mirrors a public discussion from the WebODM Discord. How to request a removal.